University of Pretoria Department of Economics Working Paper Series

Education and Fertility: Panel Evidence from sub-Saharan Africa Carolyn Chisadza University of Pretoria

Manoel Bittencourt University of Pretoria Working Paper: 2015-26 May 2015

__________________________________________________________ Department of Economics University of Pretoria 0002, Pretoria South Africa Tel: +27 12 420 2413

Education and Fertility: Panel Evidence from sub-Saharan Africa Carolyn Chisadzay

Manoel Bittencourtz

May 6, 2015

Abstract We study the e¤ects of di¤erent levels of education on fertility in 48 sub-Saharan African countries between 1970 and 2010. The results, based on panel data analysis with …xed e¤ects and instrumental variables, show how that lower education levels do not have a significant e¤ect on people’s fertility decisions. However, the results from the higher education levels suggest otherwise. They are indicative of a region that is transitioning from the Malthusian epoch to a modern growth regime in which people substitute quantity for quality of children. Lower fertility implies less strain on public expenditure, higher human capital and higher productivity which can lead to sustained economic growth as witnessed in most developed regions today. Keywords: education, fertility, sub-Saharan Africa JEL Classi…cation: O55, J13, I25

We acknowledge the comments that we received at the brown bag seminar at the University of Pretoria, 2014 and PhD workshop hosted by Economic Research Southern Africa in Pretoria, 2014. y Department of Economics, University of Pretoria, Lynnwood Road, Pretoria, 0002, RSA, email: [email protected]. Tel: +27 12 4206914. z Department of Economics, University of Pretoria, Lynnwood Road, Pretoria, 0002, RSA, email: [email protected]. Tel: +27 12 4203463.

1

1

Introduction

Does education attainment lower fertility? This question is relevant because both education and lower fertility rates are viewed as important processes of economic development within countries transitioning from developing to developed economies (Becker, Cinnirella & Woessmann 2010; Galor 2005; Reher 2011). Education especially has been linked to improved technology and skilled labour which in turn improves productivity (Becker, Murphy & Tamura 1990; Hansen & Prescott 2002). To date empirical analysis has given signi…cant attention to the contributory role that the rise in demand for education plays in lowering fertility rates and giving rise to the demographic transition and its spread to regions outside Western Europe (Becker et al. 2010; Bittencourt 2014; Doepke 2004; Galor 2005, 2012). Our study adds to this literature by examining the post-independence transition of sub-Saharan economies using the uni…ed growth theory. This theory has been cited as inducing the child quantity-quality trade-o¤ which eventually resulted in the demographic transition from high to low fertility rates. Most developed economies of today are characterised by high human capital accumulation, low fertility rates and high levels of productivity. We investigate whether the e¤ects of di¤erent levels of education, along with other associated variables suggested by literature, such as infant mortality and income per capita, induce a decline in the fertility rates of 48 sub-Saharan African countries between 1970 and 2010. Using panel data analysis with …xed e¤ects to control for heterogeneity and instrumental variables to account for endogeneity, we …nd that lower education levels do not have a signi…cant e¤ect on people’s fertility decisions. However, the results from the secondary education levels show a consistent negative relationship with fertility, suggesting that higher levels of education are signi…cant in lowering fertility in the region. This result is evidence of economies that are entering their own demographic transitions and moving from a Malthusian stagnation to modern economic growth, albeit more than a century after Western Europe (Galor 2005). Several explanations in literature have been reviewed as triggering the decline in fertility rates. Firstly, the Barro-Becker theory (1988, 1989) which focuses on opportunity costs involved with rising income per capita which

2

may induce parents to substitute the quantity of children for higher quality1 . Secondly, the uni…ed growth theory which emphasises the role of technology in encouraging investments in child education (Galor 2005)2 . Thirdly, the decrease in the gender gap which raises the cost of children (Galor & Weil 1996)3 . Fourthly, the change in traditions regarding the old-age security hypothesis which views the younger generation as a measure of security for the older generation (Galor 2012; Reher 2011)4 . Lastly, the declining mortality rates which reduces the need to have more children to replace those that may not survive (Conley et al. 2007; Murtin 2013). However, the di¤erences in the timing of the fertility declines have also given rise to the di¤erences in the take-o¤s of the demographic transitions, and this has led to the varying levels of economic development which we …nd between developed and developing economies today (Cervellati & Sunde 2013; Doepke 2004; Galor 2005; Galor & Mountford 2008; Reher 2011). Figure 1 illustrates these di¤erences within sub-Saharan Africa. The more mature economies in the region, such as Botswana, Mauritius, Seychelles and South Africa, have earlier take-o¤s in fertility declines than the poorer economies such as Democratic Republic of Congo, Eritrea, Niger and Rwanda.

1

Becker and Barro (1988, 1989) …nd that when the opportunity costs of raising children are high, either via increased wage rates or tax on children, they lower fertility in a model of intergenerational altruism. (See also Becker, Murphy & Tamura 1990). 2 According to the uni…ed growth theory, the process of development is divided into three distinct periods, the Malthusian epoch which is characterised by relatively constant income per capita and population growth, negligible technological progress and low returns on investment in education. As a result the relationship between income per capita and population growth is positive. The second period is the Post-Malthusian regime. As technological rates increase, the demand for skilled labour also increases which in turn raises the returns on human capital accumulation and encourages the population to invest in the education of their children, and have less children, a process known today as the child quantity-quality trade-o¤. This demographic transition allows income to keep rising and helps to move the economy into the third sustained growth regime characterised by low fertility rates, low population growth rates, high skilled labour, high income per capita and high productivity (Galor 2005, 2012; Galor & Weil 1999, 2000; Galor & Moav 2002). 3 Galor & Weil (1996) determine that the reduction in the gender gap has resulted in lower fertility rates. As demand for women’s participation in the labour force increases, so do the wages for women which raise the cost of children relatively more than they raise household income, hence leading to decisions to have fewer children. (See also Schultz 2008; Van der Vleuten & Kok 2014). 4 The introduction of capital markets, the establishment of national pension schemes, not to mention nursing homes, negated the traditional views of having many children for old age security (Galor 2012; Reher 2011).

3

Benin

Botswana

Burkina Faso

Burundi

Cameroon

Cape Verde

Central African Republic

Chad

Comoros

Democratic Republic of Congo

Republic of Congo

Ivory Coast

Djibouti

Equatorial Guinea

Eritrea

Ethiopia

Gabon

Gambia

Ghana

Guinea

Guinea-Bissau

Kenya

Lesotho

Liberia

Madagascar

Malawi

Mali

Mauritania

Mauritius

Mozambique

Namibia

Niger

Nigeria

Rwanda

SaoTome & Principe

Senegal

Seychelles

Sierra Leone

Somalia

South Africa

Sudan

Swaziland

Tanzania

Togo

Uganda

Zambia

Zimbabwe

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010

1970 1980 1990 2000 2010

2 4 6 8 2 4 6 8

2 4 6 8

fertility

2 4 6 8

2 4 6 8

2 4 6 8

Angola

2 4 6 8

1970 1980 1990 2000 2010

t Graphs by countries

Figure 1: Fertility rates across countries (Source: World Development Indicators, 1970-2010)

A comparison of global regions’population growth and fertility rates in Figure 2 shows a delay in the decline of sub-Saharan Africa’s rates. The population growth rates in the region decrease slightly during the late 1980s due to the improvements in post-independence child health care which increased survival of infants and led to fertility declines (Van der Vleuten & Kok 2014). Most of the regions, including other developing ones like South Asia and Latin America, are already exhibiting declining population growth rates by the late 1970s. The fertility rates for the other regions also decline over the period to between 2 and 3 children per women, while sub-Saharan Africa to date is still double that (Conley et al. 2007). This delay in the decline of fertility rates makes our paper relevant in investigating the triggers for the take-o¤, speci…cally if there is evidence of a trade-o¤ between education and fertility present in sub-Saharan Africa as postulated by the uni…ed growth theory, and what the implications of this trade-o¤ may be towards economic development in the region.

4

8 6

3

Fertility

2

4

popgr

2

1 0 1960

1970

1980

1990

2000

2010

1960

1970

1980

t

1990

2000

2010

t

East Asia

Europe

East Asia

Europe

Latin America

N. America

Latin America

N. America

South Asia

Africa

South Asia

Africa

Figure 2: Population Growth Rates and Total Fertility Rates

(Source: World

Development Indicators, 1960-2010)

2

Empirical Analysis

2.1

Data

We use a sample of 48 countries covering 41 years from 1970 to 2010. Most data for African countries is available from the year of the country’s independence, which in this case is from the 1960s onwards. Given that several European countries were already exhibiting a quantity-quality trade-o¤ in the 19th century (Becker et al. 2010; Galloway et al. 1998; Klemp & Weisdorf 2012), inclusion of countries from other regions may not give a true re‡ection of the e¤ects of education on fertility in sub-Saharan Africa. The dependent variable (fertility) is the total fertility rate which measures the number of births per woman and is obtained from the World Development Indicators (WDIs). We use three di¤erent variables for education levels. Primary education (primary educ) measures the gross primary enrollment rate as a percentage of the population. The secondary education variable (secondary educ) is the gross secondary enrollment rate as a percentage of the population. The third education variable (girl-boy educ) is a measure of the gender gap in 5

schooling. The variable is the ratio of girls to boys in primary and secondary education. All three variables are obtained from the WDIs. A negative and signi…cant coe¢ cient for the education variables suggests a trade-o¤ between education and fertility. This trade-o¤ indicates a child quantityquality preference as educated people realise the bene…ts of schooling and start investing more in the education of their o¤spring. According to the uni…ed growth theory this quantity-quality trade-o¤ played a signi…cant role in the onset of the demographic transitions in Western Europe, (Becker et al. 2010; Galor 2012). To avoid omitted variable bias we introduce some control variables based on the various literature (Becker & Barro 1988; Becker et al. 1990; Cervellati & Sunde 2013; Conley et al. 2007; Galor 2005). These controls include infant mortality and income per worker. The mortality variable (mortality) is the infant mortality rate per 1,000 live births taken from the WDIs. We expect a positive relationship between mortality and fertility rates. As fewer infants die due to improved health facilities and knowledge, fertility rates should decline (Cervellati & Sunde 2013; Conley et al. 2007; Reher 2011). Income per worker (gdp) is taken from the Penn World Tables 7.1 and is converted using the purchasing power parity at 2005 constant prices. We expect a negative relationship between income and fertility rates which may suggest that as income increases, the opportunity cost of raising children increases resulting in people choosing to have fewer children (Becker & Barro 1988; Becker et al. 1990; Galor & Weil 1996; Schultz 2008). For robustness, we also include urbanisation and con‡ict as added control variables (Galloway et al. 1998; Vandenbroucke 2004). As stated in Galloway et al. (1998), people that live in urban areas may have better access to information concerning contraception, di¤erences in perceived value of children, and a greater receptivity to newer ideas when compared to rural areas. Urban areas are also more technologically advanced which according to the uni…ed growth theory is instrumental in families ’decisions to have less children (Galor 2005). New technology creates a demand for the ability to analyse and evaluate new production possibilities, which raises the return to education (Galor & Weil 2000). As a result people have more incentive to get educated and to invest in the education of their children than rural based population. We therefore expect urbanisation to be associated with lower fertility rates. The urbanisation variable (urban) is taken from the 6

WDIs and measures urban population as a percentage of total population. Evidence by Vandenbroucke (2014), …nds that con‡ict may cause a negative shock to the household by increasing the probability of a woman remaining alone and reducing income via the death of the husband, or the period that he is away from home. This may negatively impact on fertility during the war. For instance, Caldwell (2004) provides evidence that fertility declined in several European countries during various episodes of social and political unrest. However, Vandenbroucke (2014) also …nds that fertility rebounded postwar induced by a catch-up e¤ect from households that could still have children. Since con‡ict has been persistent in sub-Saharan Africa over the years, we expect that fertility may have been adversely affected during times of unrest, but there may also be a temporary increase in birth rates during times of stability. The variable is obtained from the Armed Con‡ict and Intervention datasets (2013) compiled by the Center for Systemic Peace and it captures the number of major episodes of international con‡ict involving the country between 1970 and 2010. All variables are logged, except con‡ict.

2.2

Descriptive Statistics

We show the means, standard deviations, as well as the minimum and maximum statistics in Table 1. According to Van der Vleuten & Kok (2014), the fertility rates in the region have remained high until recently and this is shown by the average fertility rate in the region which is about 5.9 children per woman. Interestingly, when we look at the data in detail, we …nd that the richer economies, such as Botswana, Mauritius, Seychelles (with the highest income per worker at $62,338.66) and South Africa, are also characterised by lower fertility rates (Mauritius has the lowest at 1.47 children per woman), lower mortality rates (Seychelles has the lowest at 11.5 children per 1,000 births), less episodes of con‡ict and higher education attainment rates (both Mauritius and Seychelles are among the countries with the highest secondary and primary enrollment rates). The opposite holds true for the poorer countries. The Democratic Republic of Congo (DRC), Eritrea, Niger and Rwanda are some of the poorer economies (DRC has the lowest income per worker at $481.95) in the region. They are characterised by high fertility rates (Rwanda is among the highest at 8.3 children per woman), high mortality rates, more episodes of con‡ict 7

(Rwanda with the highest at 10, coming from the genocide in 1994) and low education attainment rates (Eritrea and Niger have some of the lowest primary enrollment rates). These countries’ economies have also been adversely a¤ected by con‡ict over the years, and the high fertility rates in these countries may re‡ect the catch-up e¤ect suggested by Vandenbroucke (2014).

Table 1: Descriptive Statistics Variables Fertility Primary Educ Secondary educ Girl-boy educ Mortality Gdp Urban Conflict

Obs 1945 1612 1244 1113 1924 1916 1968 1812

Mean Std. Dev. 5.89 1.23 80.27 32.95 25.28 22.35 79.19 20.32 94.27 35.35 5226.45 7343.16 30.63 15.67 0.85 1.81

Min 1.47 7.86 1.06 29.42 11.50 481.95 2.38 0

Max Sources 8.29 World Bank 207.82 World Bank 122.20 World Bank 146.83 World Bank 199.50 World Bank 62338.66 Penn World Tables 85.84 World Bank 10 Center for Systemic Peace

Figure 3 depicts the geographical distribution of fertility and gross primary enrollment rates in the region for 1989 and 2009. The distribution con…rms that the relatively mature economies, such as Botswana, Mauritius and South Africa show higher levels of education and lower fertility rates compared to the poorer economies, such as Niger and Rwanda. Data coverage also improves over the years in the region, especially primary enrollment rates, which helps with more accurate analysis of the economies.

8

Figure 3: Fertility and Primary Enrollment Rates in sub-Saharan Africa (Source: World Development Indicators) We o¤er a brief look at the correlations in Table 2. The negative correlation between the education variables and fertility suggests a trade-o¤ in education and fertility. The remaining controls are statistically in line with expectations. Infant mortality is positively correlated with fertility, as is the con‡ict variable suggesting a postwar catch-up e¤ect in the region. Both urbanisation and income per worker are negatively correlated with fertility suggesting that as countries become more developed, fertility rates start to decline. Among the determinants of fertility, secondary education, the ratio of girls to boys education and infant mortality are the most correlated with the dependent variable.

9

Table 2: Correlation Matrix Fertility Primary educ Secondary educ Girl-boy educ Mortality Gdp Urban Conflict Fertility 1.000 Primary educ -0.494* 1.000 Secondary educ -0.797* 0.576* 1.000 Girl-boy educ -0.568* 0.676* 0.530* 1.000 Mortality 0.699* -0.555* -0.728* -0.661* 1.000 Gdp -0.492* 0.332* 0.660* 0.341* -0.478* 1.000 Urban -0.546* 0.317* 0.523* 0.190* -0.438* 0.506* 1.000 Conflict 0.168* -0.124* -0.169* -0.091* 0.207* -0.129* -0.137* 1.000 Sources: World Bank, Center for Systemic Peace, Penn World Tables

When we plot the di¤erent mean education levels against mean fertility rates in Figure 4, a similar negative relationship is depicted in the graphs. This characteristic is in line with the quantity-quality trade-o¤ theory which may indicate the onset of demographic transitions within the region.

Secondary Education and Fertility Rates mean fertility 4 6 8

mean fertility 4 6 8

Primary Education and Fertility Rates Niger

Niger

Mali Congo, Dem. Rep. Burkina Faso Chad Senegal Ethiopia Cote d'Ivoire NigeriaKenya Guinea-Bissau Tanzania Madagascar Mozambique EritreaBurundi Gambia, The Mauritania Djibouti Equatorial Guinea Rep.Tome and Principe ZimbabweCongo,Sao Botswana Gabon South Africa

0

50 100 mean primary education Fitted values

South Africa Seychelles

2

2

Seychelles

Mali Burkina Faso Congo, Dem. Rep. Chad Senegal Ethiopia Cote d'Ivoire Nigeria Kenya Guinea-Bissau Tanzania Burundi Madagascar Mozambique Gambia, Eritrea The Guinea Mauritania Djibouti Equatorial Congo, Rep. Zimbabwe Sao Tome and Principe GabonBotswana

150

mfertility

0

20 40 60 mean secondary education Fitted values

80

mfertility

mean fertility 4 6 8

Gender Gap in Education and Fertility Rates Niger Mali Congo, Dem. Burkina Faso Rep. Chad Senegal Ethiopia Cote d'Ivoire Nigeria Tanzania Guinea-Bissau Kenya Burundi Madagascar Mozambique Gambia, Djibouti The Eritrea Mauritania Equatorial Guinea Congo, Rep.Sao Tome and Principe Zimbabwe Gabon

Botswana South Africa

2

Seychelles

40

60 80 100 120 mean ratio of girls to boys in education Fitted values

mfertility

Figure 4: Education and Fertility Rates in sub-Saharan Africa (Source: World Development Indicators, 1970-2010)

3

Methodology

We estimate the following equation using panel data techniques: 10

f ertilityit =

i

+

1 educit

+

i Xit

+

it

where f ertility represents the dependent variable, educ represents the three education explanatory variables entered in separate models and X is a vector of control variables. The panel data approach allows us to control for heterogeneity, as well as test for more behavioural models than purely cross section or time series. This helps us to get a more informative analysis of the region. We estimate a baseline pooled OLS (POLS) model which assumes homogeneity among the countries, that is they share common intercepts and slopes. However, countries like South Africa and Nigeria will not necessarily exhibit similar characteristics in trade policies, …scal and monetary policies, political barriers, population growth, geographic location or access to technology. The …xed e¤ects

i

capture the heterogeneity present in the model

by taking these di¤erences into account and incorporating individual speci…c e¤ects, allowing for more e¢ cient estimates. Since endogeneity, in the form of reverse causality, may be present in the model via education and mortality, we use instrumental variables (IV) to minimise this problem5 . The IV method allows consistent estimation in large samples where the explanatory variables are correlated with the error terms of a regression relationship. In other words, the instrumental variables used only in‡uence the level of fertility through their impact on education and mortality. We instrument primary education with …nancial development (credit), secondary education with globalisation (globalisation), the ratio of girls to boys education with a post cold war dummy (post-cold war ), and infant mortality with immunisation against measles (measles). 5 For instance, Becker et al. (2010) …nd that causation between fertility and education runs both ways. Higher fertility may also discourage investments in human capital. Alternatively, higher stocks of capital may reduce the demand for children because that raises the cost of the time spent on child care (Becker et al. 1990). Furthermore, Klemp & Weisdorf (2012) show that having more children in a family reduces their chances of becoming literate and skilled in the 18th -19th century England. Conley et al. (2007) highlight the question of causal directionality between child mortality and fertility rates. They argue that increased child mortality may be due to increased fertility which increases strain on household resources, decreases parental care and supervision with the addition of more children. As reported by Dreze & Murthi (2001), high fertility may raise child mortality for biological (age of giving birth) or behavioural reasons (cultural preferences for sons instead of daughters), while high child mortality may raise fertility rates by inducing parents to replace the lost children.

11

All the instrumental variables are logged, except the dummy. Finding instruments always proves a di¢ cult task in empirical analysis, however in our opinion, these instruments represent exogenous shocks to sub-Saharan countries during the post-independence period. As such, we do not expect them to in‡uence fertility rates directly, but rather to work through the various channels of education and infant mortality in reducing fertility rates. For example, if there are credit constraints making borrowing di¢ cult and costly, then access to invest in human capital is impeded, whereas easy access to credit allows people to make investment decisions in education or otherwise (Becker et al. 1990; Galor & Moav 2004; Galor & Zeira 1993). In this instance, the exclusion restriction is that the …nancial development instrument is correlated with primary education, but not with fertility rates. The instrument therefore indirectly in‡uences fertility rates through its e¤ect in increasing access to education which in turn will lower fertility rates. According to Schultz (2008), women with access to credit or who own marketable titles to lands and other assets tend to improve their welfare and that of their children by choosing education over quantity of children. The …nancial development instrument is obtained from the WDIs and measures money and quasi money as a percentage of GDP. The second instrument accounts for the latest external wave of globalisation taking place in the world including sub-Saharan Africa. The globalisation instrument is taken from a dataset compiled by Dreher (2006) and updated by Dreher, Gaston and Martens (2008). It is made up of economic, social and political globalisation which captures the international ‡ows of goods, capital, people, information and ideas. Research undertaken by Avelino, Brown & Hunter (2005) on the e¤ects of trade openness on education …nds a positive association between the two variables. Greater international interaction between people from di¤erent nations facilitates the di¤usion of ideas thus stimulating aggregate productivity (Andersen and Dalgaard 2011). Evidence by Galor and Mountford (2008) …nds that gains from trade in developing countries is concentrated in non-industrial unskilled-intensive goods, such as agricultural produce, which results in little incentive to invest in education but rather encourages further increase in population. On the other hand, gains from trade in developed economies is used to improve the specialisation of industrial skill-intensive goods which induces a rise in demand for skilled labour and leads to a gradual investment 12

in the quality of the population. We expect globalisation to have no direct bearing on fertility rates, but rather, as cited in the literature, to increase education which in turn will lower fertility rates. The third instrument accounts for the external democratic shock coming with the end of the cold war in the 1990s. The end of the cold war was accompanied by a movement towards promoting democratic institutions and economic development in third world countries through technical and …nancial assistance. One of the main aims for international assistance to developing countries is to allow poor countries to redirect their resources to programs for improving education, health and poverty. In his analysis, Dunning (2004) …nds that the end of the cold war improved the e¤ectiveness of the Western aid conditionality, whereas during the cold war, the donor’s geopolitical objectives reduced the credibility of threats to condition aid on the adoption of democratic reforms. Research by Dreher et al. (2006) investigates the link between aid and education and they …nd results in favour of aid increasing primary school enrollment. Given that the Millennium Development Goals from international organisations and donors include increasing universal primary education and promoting gender equality, the end of the cold war dummy should only a¤ect fertility through better education, suggesting the e¤ectiveness of international assistance in increasing education levels in the region. The instrument for mortality is the immunisation against measles (% of children ages 12-23 months) and is obtained from the WDIs. Measles outbreaks are still common in many developing countries, particularly in parts of Africa which are characterised by weak health infrastructures due to overpopulation, lack of vaccines, poor dissemination of health information and con‡ict (World Health Organisation, WHO; United Nations International Children’s Emergency Fund, UNICEF). In view of these facts, routine measles vaccination coverage has been selected as an indicator of progress towards achieving the fourth Millennium Development Goal to reduce child mortality in developing countries (WHO). According to literature, declines in mortality are largely driven by improvements in public health, education and adoption of technologies (Soares 2007; Reher (2011; Van der Vleuten & Kok 2014; Schultz 2008). As such, increase in health awareness through the media and government programmes, as well as increase in imported preventative medicines from abroad such as vaccines, antibiotics, 13

arti…cial contraception, mosquito nets, antiretroviral drugs, etc. has helped infant mortality rates in sub-Saharan Africa to decline over the years. Figure 5 shows the decline in sub-Saharan Africa’s infant mortality rates over the period. But it is interesting to note in the diagrams that the mortality rates start to decline earlier than the fertility rates. This suggests that the preventative measures, such as immunisations, in‡uence fertility rates

60

4.5

5

80

mean fertility 5.5

6

mean infant mortality 100 120

6.5

140

through their e¤ects in lowering infant mortality rates6 .

1960

1970

1980

1990

2000

2010

t

1960

1970

1980

1990

2000

2010

t

Figure 5: Total Fertility Rates and Infant Mortality Rates in sub-Saharan Africa (Source: World Development Indicators, 1960-2010)

4

Results

4.1

Basic Results

We report the results in Table 3 for both the pooled OLS and the …xed e¤ects models. These results are the baseline regressions which include the variables stated by literature as encouraging the quantity-quality trade-o¤ (Galor 2005). The results indicate a negative and signi…cant relationship between secondary education and fertility rates, while the primary and gender 6

Other instruments used in literature for infant mortality include adult male mortality (Galloway et al. 1998), lagged mortality (Murtin 2013); malaria ecology and percentage of population at risk of malaria (Conley et al. 2007).

14

gap education variables become positive and insigni…cant once we account for the heterogeneity in our panel and include control variables. A 10% increase in secondary education diminishes fertility by about 0.5% to 2% suggesting that investing in higher education has more e¤ect in reducing fertility than primary level or the ratio of girls to boys in schooling. Our results are in line with Bittencourt (2014), Lehr (2009) and Murtin (2013). Empirical analysis by Bittencourt (2014) …nds a negative and signi…cant relationship between secondary enrollment rates and fertility within the Southern African Development Community (SADC) region. Similarly, Lehr (2009) also …nds that secondary education is negatively related with fertility across both high and low-productivity economies, whereas primary education is positively related to fertility, more so in low-productivity economies that have not yet experienced the demographic transition. On the other hand, Murtin (2013) …nds a negative and signi…cant relationship between fertility and all three levels of education (primary, secondary and tertiary schooling), while Becker et al. (2010) …nd that primary school enrollments already had a negative impact on fertility in 19th century Prussia. Results are ambiguous for primary education and the gender gap education variable. The primary education variable is negative and signi…cant in most of the regressions. However the inclusion of …xed e¤ects and control variables appears to undermine these results, as with the results for ratio of girls to boys in schooling. The positive primary education e¤ects may act through channels that improve health, fecundity and changes in social norms of women (Lehr 2009). Educated women may have better basic knowledge on health and thus have greater fecundity. According to Ainsworth et al. (1996), a possible reason for the positive relationship may be that girls who complete only a few years of schooling are those who become pregnant and thus do not receive the full bene…t of higher education, or those that are forced by family to get married early as they will bring in income through the customary bridal price. Their research however shows that the last years of primary female schooling a¤ect fertility negatively in about half of the fourteen sub-Saharan African countries under review, while secondary education is associated with signi…cantly lower fertility across all the countries in their sample. Alternatively, the negative e¤ect of the gender gap works through several channels in reducing fertility. Increasing female education may raise a 15

woman’s age at marriage (Ainsworth et al. 1996; Galor & Weil 1996), and it may encourage women to invest in the education of their children (Ainsworth et al. 1996). Increased education also raises women’s knowledge of contraceptive methods (Dreze & Murthi 2001), and it may increase the wage that women can earn in the labour market which raises the opportunity cost of having children (Becker & Barro 1988; Becker et al. 1990; Galloway et al. 1998; Galor and Weil 1996). Infant mortality is positively and signi…cantly related to fertility rates (Cervellati & Sunde 2013; Conley et al. 2007; Murtin 2013). Survival of infants was low in the past due to adverse health conditions during childbirth, and women may have therefore spent a considerable amount of time replacing the lost children. However, with better education in health and hygiene for mothers, and improvements in health facilities, mortality rates have gradually started to decrease reducing the need to have many children (Reher 2011). Evidence by Conley et al. (2007) indicates that the infant mortality may be the most important factor in explaining declining fertility rates globally. Moreover, Cervellati and Sunde (2013) suggest that di¤erences in infant mortality may explain a substantial part of the observed di¤erences in the timing of the demographic transition across countries, as witnessed in sub-Saharan Africa’s delay in fertility declines (Figure 5). The delay may also highlight that people’s perceptions take some time to adjust (Montgomery 2000)7 . The results for income per worker are negative and sometimes signi…cant. According to the uni…ed growth theory, the increase in technological progress, not only allows income per worker to keep rising, but also raises the demand for skilled labour. This e¤ect encourages people to invest in quality rather than quantity and thus reduces fertility (Galor & Weil 1999). The negative income results are also in line with the Barro-Becker (1988, 1989) hypothesis. Rising income per worker, through increased labour opportunities for women, may increase the opportunity costs of raising children, thus lowering fertility (Galloway et al. 1998; Galor & Weil 1996).

7

The decline in infant mortality in Western Europe during the 1800s was associated at …rst with increasing fertility rates (Galor 2012). Empirical evidence by Doepke (2005) and Murphy (2010) shows that infant mortality rates were already declining before fertility rates in England and France during the 19th century.

16

Table 3: Pooled OLS and Fixed Effects FERTILITY Primary educ Secondary educ Girl-boy educ

1 POLS -0.224*** (0.010)

2 POLS

3 POLS

-0.217*** (0.010)

4 POLS -0.017** (0.007)

5 POLS

6 POLS

7 FE -0.212*** (0.029)

-0.048*** (0.007)

8 FE

9 FE

10 FE 0.019 (0.032)

-0.182*** (0.027)

11 FE

12 FE

-0.060* (0.031)

-0.568*** (0.030)

-0.031 -0.435*** 0.073 (0.022) (0.089) (0.073) Mortality 0.435*** 0.401*** 0.434*** 0.531*** 0.404*** 0.548*** (0.014) (0.018) (0.018) (0.051) (0.065) (0.053) Gdp -0.039*** -0.032*** -0.055*** 0.028 -0.00008 -0.021 (0.005) (0.006) (0.007) (0.040) (0.046) (0.045) Observations 1,590 1,223 1,093 1,540 1,183 1,059 1,590 1,223 1,093 1,540 1,183 1,059 F test 501.61*** 492.72*** 351.02*** 618.55*** 526.44*** 487.48*** 52.85*** 44.37*** 23.98*** 52.52*** 44.34*** 53.02*** R-squared 0.184 0.445 0.271 0.672 0.691 0.696 0.167 0.405 0.158 0.548 0.556 0.598 Number of i 47 47 48 47 47 48 Country FE YES YES YES YES YES YES Coefficients reported. Robust standard errors in parentheses. *** p